Traditional data analytics approaches often ignore data security and privacy issues, and are not efficient for dynamic analysis and decision-making. The parallel data and federated data provide new ideas for addressing these issues. However, how to design the analytics applications of parallel data and federated data is challenging and calls for comprehensive tool support. This paper proposes a federated parallel data platform (FPDP) that provides an end-to-end data analytics pipeline, such as virtual data generation, federation model construction, and parallel data deduction. We implement a proof-of-concept prototype and evaluate the feasibility of the platform design using a fault diagnosis case study. The evaluation results show that the proposed approach is effective and efficient to help developing trustworthy AI applications.

A Federated Parallel Data Platform for Trustworthy AI